Projects
Applied research and technical contributions.
Selected research and development projects at ETRI and Samsung Research, each with the publications, patents and other outputs it produced. The Korean speech-based model for mild cognitive impairment screening, which I developed and which was transferred to two companies in 2026, is described under Technology transfer. Graduate research is described under Research and Publications.
Research and applications
Scientific representation learning
Role: Encoder and tokenizer development, model evaluation
Science-specific encoders and tokenizers are the focus of my current work, with applications to drug discovery. I build model inputs, combine sequence representations with 3D interaction features for protein–ligand binding-affinity prediction, and evaluate how well the models generalize beyond the proteins and molecules they were trained on.
- Ongoing work; no public outputs yet.
Speech and multimodal learning for cognitive assessment
Role: Cognitive modelling, field deployment and model integration
This work combined research on Alzheimer's disease classification from picture-description speech with applied development for mild cognitive impairment (MCI) screening. I developed the graph-based and multimodal methods and was the primary developer of the Korean speech-based MCI classification model. I also integrated the model into an interactive cognitive assessment system built around a digital human, developing its user interface, real-time voice and session handling and server integration, and managing system operation.
- Two first- and corresponding-author papers in Scientific Reports (2025)
- Granted Korean patent on vision-language and graph-based cognitive assessment (KR 10-2968743, first inventor); U.S. application published
- Korean patent application on multimodal foundation models for neurological disorders (2026)
- Tablet app for MCI screening: prediction engine and field deployment (2025)
- Korean speech-based MCI model: field deployment and transfer to two companies (2026)
- IEEE ICASSP 2025 Show & Tell demonstration and DementiaBank Korean Kang Corpus (contributor)
Event-sequence learning and collaborative machine learning
Role: Temporal modelling, model development and analysis
I developed burst- and memory-aware temporal point-process models and contributed to classifiers for dynamic environments. Clinical collaborations involved model development and analysis with domain specialists, including work on treatment-choice interpretation using clinical and socioeconomic data.
- Burst and Memory-aware Transformer (Frontiers in Computational Neuroscience, 2023; first and corresponding author)
- Co-authored papers on treatment-decision modelling, predictive coding and feature-incremental classifiers (2022–2023)
- Granted Korean patent on constrained network generation (KR 10-3005917, first inventor)
- Korean and U.S. patent applications on time-series pattern and event-cluster-based prediction (pending)
Industrial machine learning at Samsung Research
Role: Staff Engineer
At Samsung Research, I developed demand and material-order forecasting models, evaluating order predictions through inventory simulations that accounted for overstock and shortages. In advertising and recommendation, I led development of a self-attention model for multi-touch advertising attribution, working with colleagues who handled the data engineering, and designed a self-supervised architecture for cross-domain recommendation.
Technology transfer
Korean speech-based classification model for mild cognitive impairment
Role: Primary developer of the classification model; lead contributor to field deployment; contributor to the technology transfer
I was the primary developer of the Korean speech-based classification model for mild cognitive impairment (MCI) screening, and led its integration into an interactive assessment system for field deployment. In 2026 the model was transferred to two companies under two technology-transfer agreements, to which I contributed as the model's developer. The deployment was a field deployment rather than a commercial service or a clinical validation, and this entry describes my part of a team project.
Patents are listed on the CV.